1 Omsk State Technical University, Omsk, Russia
|
2 Omsk State Technical University, Omsk, Russia
|
3 Omsk State Technical University, Omsk, Russia
|
4 Omsk State Technical University, Omsk, Russia
|
Abstract. A method of personality recognition by echographic parameters of the human ear has been developed on the basis of the “naive” Bayes classifier in two modes: biometric identification (EER= 0.0053) and biometric authentication (FRR= 0.0002 at FAR<= 0.0001), respectively. A device was developed for recording the biometric characteristics of the ear; a set of echographic data was collected from the ears of 75 subjects. The spectral and cepstral characteristics of the signals reflected from the ear canal were used as biometric parameters. Several window functions for constructing spectra and cepstrograms are considered. It has been established that more than 90% of “cepstral” features have a weak correlation dependence, which allows the use of a “naive” Bayesian classifier and at the same time obtaining highly accurate results of user recognition. The advantage of Bayesian classification is the possibility of robust fast learning of the identification system.
Keywords: cepstrograms, window Fourier transform, Bayes theorem, acoustic signal, pattern recognition, machine learning.